A Framework for Adaptive QoS of Web Services using Replication

For large-scale access of web services, such as news web services connected to a database, it is challenging to provide the expected QoS attributes, such as response time and availability. Further, using a single server host is not enough to provide the expected non-functional characteristics of the published web services. Especially, if the server host experience any problem, this will downgrade the reliability and the availability of the web services. Service replication is an important technique to support the non-functional properties of the published web services. In this paper, we propose a framework to adaptively meet the expected Quality of Service (QoS) attributes of web services under high number of requests using service replication. The proposed framework uses Holt’s Linear and Exponential Smoothing (HLES) time series technique to smooth and provide short term predictions of the response times and the requests arrival rates for the accessed web services. Moreover, based upon the predictions using HLES, the framework uses the Queuing Network (QN) to model the impact of the requests load on a particular web service to predict the violation of the expected response time. The QN model takes into account the different classes of the published web services. Upon a prediction of QoS violation, the framework issues a replication decision on another server host. Finally, a load balancing mechanism is used to load balancing the requests between the replicas and the original web service based on the QN model. We present the experimental results which show the effectiveness of the proposed framework.